Your agent shouldn't die when you close the lid.
DevRig gives your AI coding agents a real Linux machine that stays on. Claude Code, Codex, or Aider in a full browser terminal — with your files, your running processes, and your own model account exactly where you left them.
You're in — in line.
We open in October. You'll hear from me twice at most before then. Two quick optional questions — each is one click and genuinely shapes what gets built first:
Where do your coding agents run today?
Which agent do you run most?
Opening October 2026 · US data centers
01 — The problem
The best coding agents are tied to the most fragile machine you own.
The lid
Agents do their best work on long autonomous runs. Close the laptop, drop the Wi-Fi, or let the OS sleep, and the run dies mid-task — along with an afternoon of context.
The hand-built box
Every machine is set up by hand: the toolchain, the repo, the credentials, the tmux scrollback you're afraid to close. None of it survives a new laptop, and none of it is anywhere else.
The second job
A raw VPS runs your agent fine — after you do the provisioning, the hardening, the backups, and the 2am babysitting. The machine works. You've just become its operator.
02 — What a Rig is
A real machine. Not a sandbox.
Each Rig is a dedicated Linux environment we provision and operate, isolated at the hardware-virtualization level, that you drive entirely from the browser — or over plain SSH.
- Isolation
- A Firecracker microVM with its own kernel — hardware virtualization, not a shared-kernel container. You get root.
- Disk
- Persistent local NVMe (mirrored). Your files, installed tools, and working state survive between sessions — because nothing ends.
- Durability
- Continuous snapshots shipped off the machine, with rollback — recover from a bad agent run instead of losing an afternoon.
- Access
- Full browser terminal — TUIs, clipboard, resize — plus direct SSH when you want your own tooling. Your keys; we don't escrow them.
- Network
- Your Rig reaches us over a single outbound TLS WebSocket. No inbound ports open on anything.
- Model account
- Yours. Sign in to your own Claude, OpenAI, or API account. We never proxy, pool, or mark up model access.
- Location
- US data centers — Oregon and Virginia, on dedicated hardware. US-only at launch.
03 — How it works
Four steps, then it's just there.
Pick a plan
Your Rig is provisioned in seconds — no image building, no cloud console, no SSH key ceremony.
Open the terminal
A real Linux shell in the browser. Install your agent CLI the way you would anywhere else.
Sign in to your own model account
Claude, OpenAI, or API keys — your credentials stay on your machine, not with us.
Close the tab
Nothing pauses — the agent keeps working while you’re gone. Open the terminal later from any device, phone included, and you catch the run still in progress or already finished, with everything it did while you were away right there on screen.
04 — Pricing
Launch pricing. Whole numbers, no meter.
Always-on machines, month-to-month, on your own model account with no per-token markup. The price is the price.
Starter
$19 /month
- 2 vCPU · 6 GB RAM
- 40 GB NVMe, mirrored
- Always-on · continuous snapshots
- Browser terminal + SSH
- One agent, comfortably
Pro
$29 /month
- 4 vCPU · 12 GB RAM
- 80 GB NVMe, mirrored
- Always-on · continuous snapshots
- Browser terminal + SSH
- Two agents in parallel worktrees
Max
$79 /month
- 8 vCPU · 24 GB RAM
- 160 GB NVMe, mirrored
- Always-on · continuous snapshots
- Browser terminal + SSH
- One big machine — single heap, one big build, not two glued Pros
Launch pricing, locked for the waitlist. If prices change before we open in October, everyone on the waitlist who subscribes within 30 days of launch keeps the prices on this page for as long as their subscription stays active — and gets 25% off the first three months.
05 — The honest comparison
Where we win, and where we don't.
Against a droplet you run yourself: a $24 droplet gives you more disk than our $19 Rig. It also gives you a second job. Add the provider's daily backups and it's $31.20/month — more than Pro, with a once-a-day restore point against our continuous snapshots — before you've priced a single hour of the operating.
On memory, where agent workloads actually live: Starter is $3.17 per GB of RAM and Pro is $2.42, against roughly $5.00–8.75 across the managed development-environment field.
Against ephemeral cloud dev environments: a 4-core Codespace left running around the clock is about $265/month — and it's built to be thrown away, which is the opposite of what a long-running agent needs. Per-second platforms price the same way: a Fly machine's RAM alone is $5/GB/month and its volumes never scale to zero, so past roughly 6.5 hours of agent time a day, Pro is cheaper even granting perfect scale-to-zero.
Against hosted agents (Devin, Cursor's background agents, web-based Claude Code): those are the vendor's agent on the vendor's ephemeral compute. A Rig runs your agent CLI, on your model account, on a machine that persists — a different thing, deliberately.
Where we lose: if you enjoy operating servers, a self-managed box is cheaper and gives you more disk. And the open-source self-hosted control planes are good and free — see the FAQ for why we think most people still shouldn't run one.
06 — FAQ
The six questions a skeptic should ask.
Why not just self-host an open-source control plane on my own server?
If you already have a server you're happy operating, do that — some of the open-source control planes are genuinely good, and what you'd be paying us for is the server and the operating. DevRig has no connect-your-own-server option: we operate every machine, and that's exactly why we can promise the snapshots, the restore, and the support. We don't support servers we can't see.
Why not a $24 droplet?
A $24 droplet gives you more disk than our $19 Rig. It also gives you a second job. Add the provider's daily backups and it's $31.20 a month — with a worse restore story — before you've spent the first evening on provisioning, hardening, and the toolchain, or the first 2am on a full disk. We sell the machine with the operating included.
Do you see my code? Do you mark up model usage?
You sign in to your own Claude, OpenAI, or API account on your Rig — we never proxy, pool, or mark up model access. We're content-blind by design: routine monitoring covers infrastructure and metadata (resource usage, connectivity), not file contents. Terminal traffic passes through our control plane to reach your browser — it is relayed, not mined. And honestly: we hold administrative access to the host your Rig runs on, like every hosting provider. We disclose that instead of pretending otherwise.
What if I want to leave?
Month-to-month, cancel anytime. You have root and direct SSH, so your data leaves the same way it arrived — git push, rsync, scp. Your repos live in your own remotes and your model account is your own, so there's nothing of yours we could hold.
It's one person. Should I trust it?
DevRig is built and operated by one person, under DevRig LLC (Wyoming). No certifications yet, no support team — but also nobody between you and the person who built it. The honest mitigations: month-to-month billing, no lock-in, your code in your own remotes, your own model account, and SSH access to take everything with you at any time.
When do you open, and how does the waitlist work?
October 2026, in batches — these are real machines on real hardware, so we open in waitlist order rather than all at once. Everyone on the waitlist who subscribes within 30 days of launch keeps the prices on this page for as long as their subscription stays active, and gets 25% off the first three months. You'll get at most two emails from us before launch.
Give your agents a machine that stays on.
Opening October 2026, in waitlist order. Prices on this page, locked.